Digital Reconstruction of Analog NTSC Video

Eric Bradley

December 10, 2007

Objectives

� Obtain RF Sampled NTSC Data

� Analyze the Data / Spectrum

� Reconstruct Audio and Luminance Data

� Compare to Original Data

� Analyze Color Signal

Data Capture

� 2 Independent Sources

PoorNo60 secondsAl Jazeera

Newscast

5-11 MHz32

MS/s

Dr. Nicholas

Beser

Dataset 2

GoodYes2 seconds & 3 seconds

Aspen Ski Promo

0-6 MHz12.5 MS/s

Eric Bradley

Dataset 1

Relative Signal Quality

Source Video Available

Clip LengthVideo Content

Frequency Band

Sample Rate

Data Collector

Data Collection Systems

RF DownconverterNI PXI-5600

A/D (14-bit 100 MS/s)

NI PXI-5142

Spectrum AnalyzerAgilent E4440A

PC w/ LabView(RFSA Stream Binary

IQ Data to Disk)

VCRPanasonic PV-7451

Ch 3 RF

60-66 MHz

Baseband

0-6 MHz

12.5 MS/s3 seconds

Ch 3 RF60-66 MHz

14-bit 32 MS/s

IF (5-11 MHz)

Dataset 1 Dataset 2

IF Filter and Downconvert

RF and BasebandFrequency Response

Luminance Carrier

1.25 MHz Color Carrier4.83 MHz

Audio Carrier5.75 MHz

Audio Reconstruction

� FM Demodulation @ 4.5 MHz (above luminance)

� BPF +/- 15 KHz surrounding audio center

� Downsample to 44.1 KHz

Audio Reconstruction

Dataset 1

Dataset 2

Luminance Reconstruction

� VSB Demodulation @ 1.25 MHz

� LPF to 4.2 MHz

� Comb or Trap Filter would be better to remove Color Harmonics

� Normalize to 0-100

Luminance Reconstruction

� Field Rate: 59.94 Hz

� 262.5 lines per field

� 700 pixels per line

� 2 fields per frame

Back Porch

Front Porch

Horizontal

Sync

One HorizontalLine

Blanking LevelVertical Sync

Top of FieldBottom of Field Black Level

One Horizontal Line

Horizontal

Sync

Vertical Sync Pulses

� Algorithm not tolerant to significant noise

One Field

One Field

Dataset 1 Dataset 2

Variable Luminance Carrier Frequency in Dataset 2

� Demodulation with frequency slightly wrong (10s of Hz) results in unreadable signal

One Field

Luminance Reconstruction Results Dataset 1

Original(captured with commercial frame capture hardware)

Reconstruction

Luminance Reconstruction Results Dataset 1

Original(captured with commercial frame capture hardware)

Reconstruction

Luminance Reconstruction Results Dataset 2

Example Frame(not typical)

Color Video Reconstruction

� Chrominance I (orange/cyan)

� Chrominance Q (green/purple)

� BPF 2.1 – 4.1 MHz

� QAM Demodulation @ 3.58 MHz

� LPF 1.5 MHz (I), 0.5 MHz (Q)

� YIQ -> RGB

Color Burst

� 3.58 MHz Color Burst on Luminance to Synchronize Phase is Critical

� Not Found in Datasets - Try it Anyways!

Unsynchronized I & Q for 6 Solid Color Fields

Color Video Reconstruction

Reconstruction

Original

“Lucky” Proper Color Frame

Typical Color Frame Next Frame

Color Video Reconstruction

ReconstructionOriginal

Color Video Reconstruction

Solid Blue Background Video

Possible Algorithm Improvements

� Color Burst

� Additional Filtering – Comb or Trap

� More Robust Vertical Sync Detector and/or Normalizer

� Manually or Automatically Adjust Modulation Frequency as it Drifts for Dataset 2

Performance

� Dataset 1 – 3 seconds of audio/video

� 60 minutes of CPU time

� ~50 MB RAM

� Dataset 2 – 60 seconds of audio/video

� No vertical syncs

� 285 minutes of CPU time

� 100s MB RAM

Conclusion

� NTSC Luminance and Audio can be Reconstructed Digitally

� Clean Data is Relatively Easy

� Noise Corrupts Syncs

� Color Synchronization is Harder

� Frequency Drift is Harder

� Questions